Intel / article

Clarifying limitations of time horizon

Source: metr.org
Date: Published Jan 22, 2026

Context

An AI's time horizon is expressed in human working time, not the number of hours the AI runs unattended. In METR's explanation of the metric's limits, the 50% horizon describes the human task duration associated with a model succeeding half the time. The estimate comes from a particular task collection and can have wide uncertainty.

That makes the measure useful for tracking broad changes in capability, while leaving a separate question about delegation. A task that is costly to get wrong may require much higher reliability, and prompting, checking and correcting the still take human time. METR therefore cautions that a longer measured horizon does not translate directly into the same increase in workplace productivity.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source metr.org